Remove private reflection from OmniSharp

[333fred]

This updates OmniSharp to consume the new ExternalAccess assemblies that were added to Roslyn, providing a more structured form of internals access that will give us compile-time diagnostics and avoid the overhead of reflection-based calls. As a part of this, I'm updating our Roslyn dependency to 3.11, and I'm removing the deprecated AutoComplete handler in favor of the modern completion handlers.

[333fred]

This is in draft and tests will fail until dotnet/roslyn#53253 is merged and we can get a version with the bugfix in it, as the pickmembers service is currently failing.

[333fred]

@filipw @JoeRobich for review.

[filipw]

@filipw @JoeRobich for review.

I have a feeling it would be good to do one last release with everything we have so far before merging this

[filipw]

there is also a bunch of reflection calls here https://github.com/OmniSharp/omnisharp-roslyn/blob/master/src/OmniSharp.Script/ScriptProjectProvider.cs#L127-L140 looks like this is not part of the current change

[333fred]

there is also a bunch of reflection calls here https://github.com/OmniSharp/omnisharp-roslyn/blob/master/src/OmniSharp.Script/ScriptProjectProvider.cs#L127-L140 looks like this is not part of the current change

Unfortunately, I didn't see these when I was creating the external access library, and didn't include them in it. We can look at them as a follow up.

[333fred]

Unfortunately, I didn't see these when I was creating the external access library, and didn't include them in it. We can look at them as a follow up.

@filipw the things being used by ScriptProjectProvider are compiler APIs, not IDE APIs. ExternalAccess cannot provide IVT to them as it doesn't have IVT to them itself. Those are going to have to stay as is.

[filipw]

@filipw the things being used by ScriptProjectProvider are compiler APIs, not IDE APIs. ExternalAccess cannot provide IVT to them as it doesn't have IVT to them itself. Those are going to have to stay as is.

ah I see, makes sense!

given that we reset binder flags there and you could use it for all kinds of mad things (e.g ignore accessibility checks), I can imagine it's probably not one that will ever be exposed to anyone anyway 😅

[filipw]

thanks! looks good to me, pending a 'current' release before the merge #2150 (comment)

[333fred]

given that we reset binder flags there and you could use it for all kinds of mad things (e.g ignore accessibility checks), I can imagine it's probably not one that will ever be exposed to anyone anyway 😅

Yeah, the compiler layer gives IVT to no-one, and that's not going to be changing at any point... ever :).

[JoeRobich]

thanks! looks good to me, pending a 'current' release before the merge

Release created - https://github.com/OmniSharp/omnisharp-roslyn/releases/tag/v1.37.9